Thera Bank Personal Loan Campaign¶
Data Description:¶
The dataset contains data on 5000 customers. The data include customer demographic information (age, income, etc.), the customer's relationship with the bank (mortgage, securities account, etc.), and the customer response to the last personal loan campaign (Personal Loan). Among these 5000 customers, only 480 (= 9.6%) accepted the personal loan that was offered to them in the earlier campaign.
Context:¶
This case is about a bank (Thera Bank) whose management wants to explore ways of converting its liability customers to personal loan customers (while retaining them as depositors). A campaign that the bank ran last year for liability customers showed a healthy conversion rate of over 9% success. This has encouraged the retail marketing department to devise campaigns with better target marketing to increase the success ratio with a minimal budget.
Attribute Information:¶
- ID: Customer ID
- Age: Customer's age in completed years
- Experience: #years of professional experience
- Income: Annual income of the customer (\$000)
- ZIP Code: Home Address ZIP
- Family: Family size of the customer
- CCAvg: Avg. spending on credit cards per month (\$000)
- Education: Education Level. 1: Undergrad; 2: Graduate; 3: Advanced/Professional
- Mortgage: Value of house mortgage if any. (\$000)
- Personal Loan: Did this customer accept the personal loan offered in the last campaign?
- Securities Account: Does the customer have a securities account with the bank?
- CD Account: Does the customer have a certificate of deposit (CD) account with the bank?
- Online: Does the customer use internet banking facilities?
- Credit card: Does the customer use a credit card issued by the bank?
Learning Outcomes:¶
- Exploratory Data Analysis
- Preparing the data to train a model
- Training and making predictions using a classification model
- Model evaluation
Objective:¶
The classification goal is to predict the likelihood of a liability customer buying personal loans.
Steps and tasks:¶
- Import the datasets and libraries, check datatype, statistical summary, shape, null values or incorrect imputation. (5 marks)
- EDA: Study the data distribution in each attribute and target variable, share your findings (20 marks)
- Number of unique in each column?
- Number of people with zero mortgage?
- Number of people with zero credit card spending per month?
- Value counts of all categorical columns.
- Univariate and Bivariate
- Get data model ready
- Split the data into training and test set in the ratio of 70:30 respectively (5 marks)
- Use the Logistic Regression model to predict whether the customer will take a personal loan or not. Print all the metrics related to evaluating the model performance (accuracy, recall, precision, f1score, and roc_auc_score). Draw a heatmap to display confusion matrix (15 marks)
- Find out coefficients of all the attributes and show the output in a data frame with column names? For test data show all the rows where the predicted class is not equal to the observed class. (10 marks)
- Give conclusion related to the Business understanding of your model? (5 marks)
In [1]:
# Importing libraries
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
import seaborn as sns
import plotly.express as px
from sklearn.linear_model import LogisticRegression
logit = LogisticRegression(solver='newton-cg')
from sklearn.model_selection import train_test_split as tts
from sklearn.impute import SimpleImputer as si
from sklearn import metrics as mtr
import statsmodels.api as sm
In [2]:
import warnings
warnings.filterwarnings('ignore')
pd.options.display.float_format = '{:,.2f}'.format
from IPython.core.display import display, HTML
display(HTML("<style>.container { width:90% !important; }</style>"))
sns.set_style(style='darkgrid')
In [3]:
# Reading file
df = pd.read_csv('Bank_Personal_Loan_Modelling.csv')
# Renaming columns for ease of working with them
df.rename(columns={'ZIP Code':'Zipcode','Personal Loan':'AccptPersLoan','Securities Account':'HaveSecAcct','CD Account':'HaveCDAcct','Online':'HaveOnline','CreditCard':'HaveCC'},inplace=True)
#Checking data-types
df.info()
Inference: Non-null 5000 row data-set with integer or float data-types¶
In [4]:
df.shape
Out[4]:
In [5]:
# Checking for null values
df.isnull().sum()
Out[5]:
Inference: No Null values in the data-set¶
In [6]:
#Checking for 'zero' values
df[df==0].count()
Out[6]:
Inference: There are a number of 'zero' values. Some of these are valid data - in Mortgage, Personal Loan, Securities Account, CD Account, Online, CreditCard.
Need to investigate if 'zero' in Experience and in 'CCAvg' is valid or if they need to be imputed.
In [7]:
#Statistical Summary
df.describe().T
Out[7]:
Inference:¶
- ID seems to be unique, no duplicates
- Experience has a 'negative' value - which implies data issues, which need to be resolved
- Mortgage column has outliers that need to be addressed
- Experience, Zipcode, Education, Personal Loan, Securities Account, CD Account, Online, CreditCard are all categorical variables
- Zipcode has data-entry issues with a 4 digit zipcode showing as min value
In [8]:
# Checking for skewness of the data
df.skew()
Out[8]:
In [9]:
# Visual analysis of data skewness
df[['Age','Experience','Income','Zipcode','Family','CCAvg','Education','Mortgage']].hist(bins=25,figsize=[25,15])
Out[9]:
Inference:¶
- Zipcode with that 1 incorrect data is skewed to the left by a large extent
- CCAvg is skewed right
- Income is skewed right
- Rest of the continous values seem to be normally distributed
EDA¶
- Number of unique in each column?
- Number of people with zero mortgage?
- Number of people with zero credit card spending per month?
- Value counts of all categorical columns.
- Univariate and Bivariate
- Get data model ready
In [10]:
# Number of unique in each column?
df.nunique()
Out[10]:
In [11]:
# Number of people with zero mortgage?
print('Number of people with zero mortgage : ',df.ID[df.Mortgage == 0].count())
In [12]:
# Number of people with zero credit card spending per month?
print('Number of people with zero credit card spending per month : ',df.ID[df.CCAvg == 0].count())
In [13]:
# Value counts of all categorical columns.
for col in ['AccptPersLoan','HaveSecAcct','HaveCDAcct','HaveOnline','HaveCC','Family','Education','Experience','Zipcode'] :
print('Value Counts of {} \n{}\n\n'.format(col,df[col].value_counts()))
Univariate and Bivariate¶
In [14]:
dfviz = df[['Age','Experience','Income','CCAvg','Mortgage','AccptPersLoan']].copy()
plt.figure(figsize=[30,20])
sns.set(font_scale=1.15)
# sns.pairplot(dfviz)
sns.pairplot(dfviz, hue='AccptPersLoan')
plt.show()
In [15]:
plt.figure(figsize=[20,15])
sns.set(font_scale=1.15)
sns.heatmap(df.corr(),annot=True,vmin=-1,vmax=1,center=0, linewidth=0.2, fmt='.2f')
plt.show()
Inference:¶
- Obviously Age and Experience have a very strong positive linear relationship
- Income and CCAvg have a positive linear relationship (as expected)
- With regards to Accepting Personal Loan, from this visualition and with the data included it appears Income and CCAvg seems to be having a positive linear relationship
- Meaning those who have higher income and having higher CCAvg seems to be accepting personal loans (habitual?!)
In [16]:
corr = df.drop('AccptPersLoan',axis=1).corr()
plt.figure(figsize=[20,15])
sns.set(font_scale=1.15)
sns.heatmap(corr,annot=True,vmin=-1,vmax=1,center=0, linewidth=0.2, fmt='.2f', cmap='coolwarm')
plt.show()
Inference:¶
- Comparing the features (dropping the Dependent variable), the positive linear relationship shows but not very strong
- Interestingly Education and CCAvg have a negative linear relationship! (maybe data issues or outliers)
- Similartly Eduation and Incomg have a negative linear relationship! (maybe data issues or outliers)
In [17]:
# Comparing the distribution of Age in relation to acceptance of Loan
fig = px.violin(df, color='AccptPersLoan', y='Experience', width=1000)
fig.show()
Inference: Given the shape of the violin is similar between the two categories, there isn't any direct inference if age influences acceptance of the loans¶
In [18]:
fig = px.violin(df, color='AccptPersLoan', y='CCAvg', width=1000)
fig.show()
Inference: Individuals with lower CC Average tend to NOT accept personal loans¶
Getting Data-set ready¶
In [19]:
# Dropping ID column since that is an identifier and not a feature
df.drop('ID', axis=1, inplace=True)
In [20]:
# Dropping observations with incorrect data (with outlier 4 digit Zipcode)
df.drop(df[df.Zipcode < 90000].index, inplace=True)
In [21]:
# Converting zipcode to bins
ziplabels = ['900-905','905-910','910-915','915-920','920-925','925-930','930-935','935-940','940-945','945-950','950-955','955-960','960+']
df['Zipcode'] = pd.cut(df.Zipcode, bins=[90000, 90500, 91000, 91500, 92000, 92500, 93000, 93500, 94000, 94500, 95000, 95500, 96000,100000], labels=ziplabels)
df.head()
Out[21]:
In [22]:
# Checking for acceptance of loans across the zipcodes
fig = px.histogram(df, x='Zipcode', color='AccptPersLoan', barmode='group',
category_orders={'Zipcode': ziplabels},
width=1000)
fig.show()
In [23]:
# Dropping observations with incorrect data (with Experience less than zero)
df.drop(df[df.Experience < 0 ].index, inplace=True)
In [24]:
# Checking distribution of CC Average
fig = px.histogram(df, x='CCAvg', marginal='rug', width=1000)
fig.show()
In [25]:
# Converting CC Average to bins - given it is already represented in '000s'
cclabels = ['0-1K','1-2K','2-3K','3-4K','4-5K','5-8K','8K+']
df['CCAvg'] = pd.cut(df.CCAvg, bins=[0, 1, 2, 3, 4, 5, 8,12], labels=cclabels)
df.head()
Out[25]:
In [26]:
df.Family = df.Family.astype('category')
df.Education = df.Education.astype('category')
In [27]:
df = pd.get_dummies(df)
In [28]:
df.describe().T
Out[28]:
In [29]:
df.info()
Split the data into training and test set in the ratio of 70:30 respectively (5 marks)¶
In [30]:
dfy = df[['AccptPersLoan']].copy()
dfX = df.drop('AccptPersLoan',axis=1).copy()
In [31]:
X_train, X_test, y_train, y_test = tts(dfX, dfy, test_size=0.3, random_state=1)
In [32]:
X_test.describe().T
Out[32]:
Modeling¶
Use the Logistic Regression model to predict whether the customer will take a personal loan or not.
Print all the metrics related to evaluating the model performance (accuracy, recall, precision, f1score, and roc_auc_score).
Draw a heatmap to display confusion matrix (15 marks)
Iterarion 0¶
In [33]:
logit.fit(X_train,y_train)
print('Training score : ',logit.score(X_train, y_train))
logit.fit(X_test,y_test)
print('Test score : ',logit.score(X_test, y_test))
In [34]:
y_predict = logit.predict(X_test)
cmat = mtr.confusion_matrix(y_test,y_predict, labels=[1,0])
dfcm = pd.DataFrame(cmat, index=['1','0'],columns=['1','0'])
plt.figure(figsize=[5,3])
sns.set(font_scale=1.2)
fig = sns.heatmap(dfcm, annot=True, fmt='d',linewidth=0.5, cbar=False)
plt.tick_params(axis='both', which='major', labelbottom = False, bottom=False, top = False, labeltop=True)
plt.ylabel('Predicted\n')
plt.show()
In [35]:
npcm = np.array(cmat)
print('Accuracy of the model : ',npcm.trace()/npcm.sum())
print('Recall of the model : ',npcm[0,0]/npcm[:,0].sum())
print('Precision of the model : ',npcm[0,0]/npcm[0].sum())
In [36]:
appr0 = np.array([[npcm.trace()/npcm.sum()],[npcm[0,0]/npcm[:,0].sum()],[npcm[0,0]/npcm[0].sum()]])
appr0
Out[36]:
Iteration 1¶
In [37]:
df1 = df.copy()
In [38]:
fig = px.histogram(df1[df1.Mortgage > 0 ], x='Mortgage', marginal='rug', width=1000)
fig.show()
In [39]:
# Converting Mortgage to bins
mortgagelabels = ['Zero','0-100K','100-150K','150-200K','200-250K','250-300K','300-400K','400K+']
df1.Mortgage = pd.cut(df1.Mortgage, bins=[-1,0,100, 150, 200, 250, 300, 400, 1000], labels=mortgagelabels)
df1.Mortgage.value_counts(normalize=True)*100
Out[39]:
In [40]:
df1 = pd.get_dummies(df1)
In [41]:
df1.HaveSecAcct = df1.HaveSecAcct.astype('category')
df1.HaveCDAcct = df1.HaveCDAcct.astype('category')
df1.HaveOnline = df1.HaveOnline.astype('category')
df1.HaveCC = df1.HaveCC.astype('category')
In [42]:
df1.info()
In [43]:
dfy1 = df1[['AccptPersLoan']].copy()
dfX1 = df1.drop('AccptPersLoan',axis=1).copy()
In [44]:
X1_train, X1_test, y1_train, y1_test = tts(dfX1, dfy1, test_size=0.3, random_state=30)
In [45]:
y1_train.AccptPersLoan.value_counts(normalize=True)*100
Out[45]:
In [46]:
logit.fit(X1_train,y1_train)
print('Training score : ',logit.score(X1_train, y1_train))
logit.fit(X1_test,y1_test)
print('Test score : ',logit.score(X1_test, y1_test))
In [47]:
y1_predict = logit.predict(X1_test)
cmat1 = mtr.confusion_matrix(y1_test,y1_predict, labels=[1,0])
dfcm1 = pd.DataFrame(cmat1, index=['1','0'],columns=['1','0'])
plt.figure(figsize=[5,3])
sns.set(font_scale=1.2)
fig = sns.heatmap(dfcm1, annot=True, fmt='d',linewidth=0.5, cbar=False)
plt.tick_params(axis='both', which='major', labelbottom = False, bottom=False, top = False, labeltop=True)
plt.ylabel('Predicted\n')
plt.show()
In [48]:
npcm1 = np.array(cmat1)
print('Accuracy of the model : ',npcm1.trace()/npcm1.sum())
print('Recall of the model : ',npcm1[0,0]/npcm1[:,0].sum())
print('Precision of the model : ',npcm1[0,0]/npcm1[0].sum())
In [49]:
appr1 = np.array([[npcm1.trace()/npcm1.sum()],[npcm1[0,0]/npcm1[:,0].sum()],[npcm1[0,0]/npcm1[0].sum()]])
appr1
Out[49]:
Iteration 2¶
In [50]:
df2 = df1.copy()
In [51]:
df2['Income_Log'] = np.log(df2.Income)
fig = px.box(df2, x='Income', width=1000, height=200)
fig.show()
fig = px.box(df2, x='Income_Log', width=1000, height=200)
fig.show()
In [52]:
df2.drop('Income',axis=1,inplace=True)
dfy2 = df2[['AccptPersLoan']].copy()
dfX2 = df2.drop('AccptPersLoan',axis=1).copy()
In [53]:
X2_train, X2_test, y2_train, y2_test = tts(dfX2, dfy2, test_size=0.3, random_state=30)
In [54]:
y2_train.AccptPersLoan.value_counts(normalize=True)*100
Out[54]:
In [55]:
logit.fit(X2_train,y2_train)
print('Training score : ',logit.score(X2_train, y2_train))
logit.fit(X2_test,y2_test)
print('Test score : ',logit.score(X2_test, y2_test))
In [56]:
y2_predict = logit.predict(X2_test)
cmat2 = mtr.confusion_matrix(y2_test,y2_predict, labels=[1,0])
dfcm2 = pd.DataFrame(cmat2, index=['1','0'],columns=['1','0'])
plt.figure(figsize=[5,3])
sns.set(font_scale=1.2)
fig = sns.heatmap(dfcm2, annot=True, fmt='d',linewidth=0.5, cbar=False)
plt.tick_params(axis='both', which='major', labelbottom = False, bottom=False, top = False, labeltop=True)
plt.ylabel('Predicted\n')
plt.show()
In [57]:
npcm2 = np.array(cmat2)
print('Accuracy of the model : ',npcm2.trace()/npcm2.sum())
print('Recall of the model : ',npcm2[0,0]/npcm2[:,0].sum())
print('Precision of the model : ',npcm2[0,0]/npcm2[0].sum())
In [58]:
appr2 = np.array([[npcm2.trace()/npcm2.sum()],[npcm2[0,0]/npcm2[:,0].sum()],[npcm2[0,0]/npcm2[0].sum()]])
appr2
Out[58]:
Iteration 3 (using Logit models with class_weight)¶
In [59]:
df3 = df2.copy()
In [60]:
dfy3 = df3[['AccptPersLoan']].copy()
dfX3 = df3.drop('AccptPersLoan',axis=1).copy()
In [61]:
X3_train, X3_test, y3_train, y3_test = tts(dfX3, dfy3, test_size=0.3, random_state=30)
In [62]:
y3_train.AccptPersLoan.value_counts(normalize=True)*100
Out[62]:
In [63]:
logit = LogisticRegression(solver='newton-cg', class_weight='balanced')
logit.fit(X3_train,y3_train)
print('Training score : ',logit.score(X3_train, y3_train))
logit.fit(X3_test,y3_test)
print('Test score : ',logit.score(X3_test, y3_test))
In [64]:
y3_predict = logit.predict(X3_test)
cmat3 = mtr.confusion_matrix(y3_test,y3_predict, labels=[1,0])
dfcm3 = pd.DataFrame(cmat3, index=['1','0'],columns=['1','0'])
plt.figure(figsize=[5,3])
sns.set(font_scale=1.2)
fig = sns.heatmap(dfcm3, annot=True, fmt='d',linewidth=0.5, cbar=False)
plt.tick_params(axis='both', which='major', labelbottom = False, bottom=False, top = False, labeltop=True)
plt.ylabel('Predicted\n')
plt.show()
In [65]:
npcm3 = np.array(cmat3)
print('Accuracy of the model : ',npcm3.trace()/npcm3.sum())
print('Recall of the model : ',npcm3[0,0]/npcm3[:,0].sum())
print('Precision of the model : ',npcm3[0,0]/npcm3[0].sum())
In [66]:
appr3 = np.array([[npcm3.trace()/npcm3.sum()],[npcm3[0,0]/npcm3[:,0].sum()],[npcm3[0,0]/npcm3[0].sum()]])
appr3
Out[66]:
Summary:¶
Comparing the results of the iterations:
In [67]:
compare = pd.DataFrame('',index=['Accuracy','Recall','Precision'],columns=['Iteration0','Iteration1','Iteration2','Iteration3'])
In [68]:
compare.Iteration0 = appr0
compare.Iteration1 = appr1
compare.Iteration2 = appr2
compare.Iteration3 = appr3
pd.options.display.float_format = '{:,.4f}'.format
compare
Out[68]:
In [69]:
cmpr = compare.T
cmpr = compare.reset_index()
cmpr = cmpr.melt(id_vars='index',value_vars=['Iteration0','Iteration1','Iteration2','Iteration3'], var_name='Iteration',value_name='Value')
# cmpr
fig = px.line(cmpr, x='Iteration', y='Value', color='index', range_y=[0.5,1.1], width=1000, height=600)
fig.data[0].update(mode='markers+lines')
fig.data[1].update(mode='markers+lines')
fig.data[2].update(mode='markers+lines')
fig.show()
Comparing...¶
- In this business context of the need to maximize the conversion (correctly identifying the individuals most likely to accept loans), it is preferable to have higher True Positives
- Meaning, it is preferable to reach out to the customers who are likely to accept personal loans
...hence the model with the 'balanced' class_weight is the right fit for this usecase with the True Positive of 124 and Precision of 0.9688¶
Find out coefficients of all the attributes and show the output in a data frame with column names?
For test data show all the rows where the predicted class is not equal to the observed class. (10 marks)
In [70]:
# Attributes and Co-efficients
dfcoef = pd.DataFrame({'Attributes':list(X3_test.columns),'Coefficient':list(logit.coef_[0])})
dfcoef.sort_values('Coefficient', ascending=False)
Out[70]:
In [71]:
# Showing predicted values different to the Actual values
y3_predict = logit.predict(X3_test)
dfpred = pd.DataFrame({'Actual':(list(y3_test.AccptPersLoan)),'Predicted':list(y3_predict)})
dfpred[dfpred.Actual != dfpred.Predicted]
Out[71]:
In [72]:
# Summary showing predicted values different to the Actual values
dfpred[dfpred.Actual != dfpred.Predicted].groupby(['Actual','Predicted'])['Predicted'].agg('count')
Out[72]:
Give conclusion related to the Business understanding of your model? (5 marks)¶
In [73]:
dfcoef.sort_values('Coefficient', ascending=False).head()
Out[73]:
In [74]:
dfcoef.sort_values('Coefficient').head()
Out[74]:
In [75]:
plt.figure(figsize=[5,3])
sns.set(font_scale=1.2)
fig = sns.heatmap(dfcm3, annot=True, fmt='d',linewidth=0.5, cbar=False)
plt.tick_params(axis='both', which='major', labelbottom = False, bottom=False, top = False, labeltop=True)
plt.ylabel('Predicted\n')
plt.show()
Based on the above data-points:¶
Ranking of factors which contribute to customers more likely to accept Personal Loans¶
- Higher Income - possibly the ability to repay loans quickly
- Having a 'Certificate of Deposit' Account (CD Account) - maybe having a safetynet in that deposit to repay for the personal loan
- Average CC balance between \$3,000 to \\$4000
- Having a large family - More needs than a smaller family
- Higher the education (Professional/Graduate) - maybe better understanding of how personal loan works & also the need for additional funds to complete their education!
Ranking of factors which contribute to customers NOT likely to accept Personal Loans¶
- Lower education seems to be the biggest factor (given the data) that negates customers taking loans
- Having smaller family - maybe there aren't unplanned expenses hitting them and hence not likely to take loans
- Having a Securities Account
- Having a CC with Average balance between \$1,000 to \\$2000
Based on the confusion matrix,¶
- The model is built for maximizing the 'True Positives', aka maximizing the return on the effort/investments in campaigns to offer personal loans to the right segment of customers
- There is also the lost opportunity with some customers who are likely to accept loans, who has been classified by this model as not likely to accept loans.
- Depending on the Bank's needs & priority, the model can be adjusted, thresholds adjusted to tilt the model to maximize, 'True Positives' or to minimize 'False Positives' and 'False Negatives'